unreal.LearningAgentsPolicy

class unreal.LearningAgentsPolicy(outer: Object | None = None, name: Name | str = 'None')

Bases: LearningAgentsManagerListener

A policy that maps from observations to actions.

C++ Source:

• Plugin: LearningAgents

• Module: LearningAgents

• File: LearningAgentsPolicy.h

Editor Properties: (see get_editor_property/set_editor_property)

• decoder_network (LearningAgentsNeuralNetwork): [Read-Only] The underlying decoder neural network.

• encoder_network (LearningAgentsNeuralNetwork): [Read-Only] The underlying encoder neural network.

• interactor (LearningAgentsInteractor): [Read-Only] The agent interactor this policy is associated with.

• is_setup (bool): [Read-Only] True if this object has been setup. Otherwise, false.

• manager (LearningAgentsManager): [Read-Only] The manager this object is associated with.

• policy_network (LearningAgentsNeuralNetwork): [Read-Only] The underlying policy neural network.

• visual_logger_objects (Map[Name, LearningAgentsVisualLoggerObject]): [Read-Only] The visual logger objects associated with this listener.

decode_and_sample_actions(action_noise_scale=1.000000) → None

Decodes and samples action vectors using the Decoder network. This should be called after EvaluatePolicy and before PerformActions.

Parameters:

action_noise_scale (float) – Scale of the action noise to use during sampling. Set this to zero to always sample the mean (expected) action.

encode_observations() → None

Encodes the buffered observation vectors using the Encoder network. This should be called after GatherObservations and before EvaluatePolicy.

evaluate_policy() → None

Calling this function will run the underlying neural network on the previously encoded observations to populate the encoded actions. This should be called after EncodeObservations and before DecodeAndSampleActions.

get_decoder_network_asset() → LearningAgentsNeuralNetwork

Gets the current Encoder Network Asset being used

Return type:

LearningAgentsNeuralNetwork

get_encoder_network_asset() → LearningAgentsNeuralNetwork

Gets the current Encoder Network Asset being used

Return type:

LearningAgentsNeuralNetwork

get_memory_state(agent_id=-1) → Array[float]

Gets the current memory state for a given agent as represented by an abstract vector learned by the policy.

Parameters:

agent_id (int32) – The AgentId to get the memory state of

Returns:

out_memory_state (Array[float]): The output memory state of the agent

Return type:

Array[float]

get_memory_state_size() → int32

Gets the size of the memory state

Return type:

int32

get_policy_network_asset() → LearningAgentsNeuralNetwork

Gets the current Policy Network Asset being used

Return type:

LearningAgentsNeuralNetwork

classmethod make_policy(manager, interactor, class_=None, name='Policy', encoder_neural_network_asset=None, policy_neural_network_asset=None, decoder_neural_network_asset=None, reinitialize_encoder_network=True, reinitialize_policy_network=True, reinitialize_decoder_network=True, policy_settings=[], seed=1234) → LearningAgentsPolicy

Constructs this object to be used with the given agent interactor and policy settings.

Parameters:

• manager (LearningAgentsManager) – The input Manager

• interactor (LearningAgentsInteractor) – The input Interactor component

• class (type(Class)) – The policy class

• name (Name) – The policy name

• encoder_neural_network_asset (LearningAgentsNeuralNetwork) – Optional Encoder Network Asset to use. If not provided, asset is empty, or bReinitializeEncoderNetwork is set then a new neural network object will be created.

• policy_neural_network_asset (LearningAgentsNeuralNetwork) – Optional Policy Network Asset to use. If not provided, asset is empty, or bReinitializePolicyNetwork is set then a new neural network object will be created according to the given PolicySettings.

• decoder_neural_network_asset (LearningAgentsNeuralNetwork) – Optional Decoder Network Asset to use. If not provided, asset is empty, or bReinitializeDecoderNetwork is set then a new neural network object will be created.

• reinitialize_encoder_network (bool) – If to reinitialize the encoder network

• reinitialize_policy_network (bool) – If to reinitialize the policy network

• reinitialize_decoder_network (bool) – If to reinitialize the decoder network

• policy_settings (LearningAgentsPolicySettings) – The policy settings to use on creation of a new policy network

• seed (int32) – Random seed to use for initializing network weights and policy sampling

Return type:

LearningAgentsPolicy

run_inference(action_noise_scale=1.000000) → None

Calls GatherObservations, EncodeObservations, EvaluatePolicy, DecodeAndSampleActions, PerformActions

Parameters:

action_noise_scale (float) –

set_memory_state(agent_id=-1, memory_state) → None

Sets the current memory state for a given agent as represented by an abstract vector learned by the policy.

Parameters:

• agent_id (int32) – The AgentId to set the memory state of

• memory_state (Array[float]) – The input memory state of the agent

setup_policy(manager, interactor, encoder_neural_network_asset=None, policy_neural_network_asset=None, decoder_neural_network_asset=None, reinitialize_encoder_network=True, reinitialize_policy_network=True, reinitialize_decoder_network=True, policy_settings=[], seed=1234) → None

Initializes this object to be used with the given agent interactor and policy settings.

Parameters:

• manager (LearningAgentsManager) – The input Manager

• interactor (LearningAgentsInteractor) – The input Interactor component

• encoder_neural_network_asset (LearningAgentsNeuralNetwork) – Optional Encoder Network Asset to use. If not provided, asset is empty, or bReinitializeEncoderNetwork is set then a new neural network object will be created.

• policy_neural_network_asset (LearningAgentsNeuralNetwork) – Optional Policy Network Asset to use. If not provided, asset is empty, or bReinitializePolicyNetwork is set then a new neural network object will be created according to the given PolicySettings.

• decoder_neural_network_asset (LearningAgentsNeuralNetwork) – Optional Decoder Network Asset to use. If not provided, asset is empty, or bReinitializeDecoderNetwork is set then a new neural network object will be created.

• reinitialize_encoder_network (bool) – If to reinitialize the encoder network

• reinitialize_policy_network (bool) – If to reinitialize the policy network

• reinitialize_decoder_network (bool) – If to reinitialize the decoder network

• policy_settings (LearningAgentsPolicySettings) – The policy settings to use on creation of a new policy network

• seed (int32) – Random seed to use for initializing network weights and policy sampling